A stronger structure theorem for excluded topological minors

Grohe and Marx proved that if G does not contain H as a topological minor, then there exist constants g=O(|V(H)|^4), D and t depending only on H such that G is a clique sum of graphs that either contain at most t vertices of degree greater than D or almost embed in some surface of genus at most g. We strengthen this result, giving a more precise description of the latter kind of basic graphs of the decomposition - we only allow graphs that (almost) embed in ways that are impossible for H (similarly to the structure theorem for minors, where only graphs almost embedded in surfaces in that H does not embed are allowed). This enables us to give structural results for graphs avoiding a fixed graph as an immersion and for graphs with bounded infinity-admissibility.

💡 Research Summary

The paper revisits the structural description of graphs that exclude a fixed graph H as a topological minor. Grohe and Marx previously showed that any such graph G can be decomposed as a clique‑sum of pieces that either contain at most t vertices of degree larger than a constant D, or “almost embed’’ into a surface of genus at most g=O(|V(H)|⁴). Their theorem, however, places no restriction on the surfaces used for the almost‑embedding pieces; consequently, a piece may be embedded in a surface that also admits an embedding of H, which is unnecessary for a topological‑minor‑free class.

The authors strengthen this result by imposing a surface‑restriction that mirrors the classic Robertson–Seymour minor structure theorem. For a given H they define the smallest genus surface S_H that does not admit an embedding of H. The main theorem then asserts that if G excludes H as a topological minor, G can still be expressed as a clique‑sum of pieces, but each almost‑embedding piece must now be placed on S_H (or a surface of even smaller genus). Moreover, the number of apex vertices and the width of vortices in each piece are bounded by constants that depend only on H, not on G. The genus bound g remains O(|V(H)|⁴), preserving the quantitative strength of the original theorem while sharpening its qualitative aspect.

To achieve this, the authors refine the tangle‑based decomposition machinery used by Grohe–Marx. They first locate a large‑order tangle in G, then apply a surface‑separator theorem that guarantees a separation of the graph into a part that can be drawn on a surface of bounded genus and a remainder of bounded tree‑width. By carefully choosing the separator they ensure that the surface obtained is a sub‑surface of S_H, i.e., a surface on which H cannot be embedded. The remainder is then handled with the standard “almost‑embedding’’ machinery: a bounded number of apex vertices, a bounded number of vortices of bounded path‑width, and a bounded number of edges crossing the surface. The key technical contribution is a “vortex compression’’ lemma that reduces the width of each vortex to a constant depending only on H, and an “apex‑control’’ argument that limits the total number of apex vertices.

Two significant applications are presented. First, the authors consider graphs that avoid a fixed graph H as an immersion. Immersion exclusion is stricter than topological‑minor exclusion, and the refined structure theorem yields a decomposition where each piece has a very limited number of apex vertices and vortices, enabling new algorithmic results for immersion‑free classes. Second, they study graphs of bounded infinity‑admissibility (a parameter measuring the size of the largest set that can be “admitted’’ by a vertex ordering). They prove that any graph with infinity‑admissibility at most k also admits a decomposition of the type described above, with constants depending on k and H. This bridges the gap between structural graph theory and parameterized algorithm design, showing that bounded admissibility forces a graph to be “almost planar’’ on a surface that forbids H.

In the concluding discussion the authors note that their theorem brings the topological‑minor theory closer to the minor theory in terms of the precision of the structural description. The dependence on |V(H)|⁴ for the genus is unchanged, but the qualitative restriction on the embedding surface is new. They suggest several directions for future work: tightening the genus bound, extending the approach to other containment notions (e.g., shallow minors), and exploiting the refined decomposition for faster algorithms on topological‑minor‑free graph classes. Overall, the paper delivers a more nuanced structural picture of topological‑minor‑free graphs, opening the door to both deeper theoretical insights and practical algorithmic applications.